#include <uapi/linux/lustre/lustre_param.h>
#include <lustre_quota.h>
#include <lustre_swab.h>
+#include <lustre_lmv.h>
#include <obd.h>
#include <obd_support.h>
#include <lustre_barrier.h>
struct obd_statfs *osfs;
struct mdt_body *reqbody = NULL;
struct mdt_statfs_cache *msf;
+ ktime_t kstart = ktime_get();
int rc;
ENTRY;
osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
}
if (rc == 0)
- mdt_counter_incr(req, LPROC_MDT_STATFS);
+ mdt_counter_incr(req, LPROC_MDT_STATFS,
+ ktime_us_delta(ktime_get(), kstart));
out:
mdt_thread_info_fini(info);
RETURN(rc);
RETURN(rc);
}
-int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
- struct md_attr *ma, const char *name)
+int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
+ struct md_attr *ma, const char *name)
{
struct md_object *next = mdt_object_child(o);
struct lu_buf *buf = &info->mti_buf;
return rc;
}
+int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
+ struct md_attr *ma, const char *name)
+{
+ int rc;
+
+ if (!info->mti_big_lmm) {
+ OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
+ if (!info->mti_big_lmm)
+ return -ENOMEM;
+ info->mti_big_lmmsize = PAGE_SIZE;
+ }
+
+ if (strcmp(name, XATTR_NAME_LOV) == 0) {
+ ma->ma_lmm = info->mti_big_lmm;
+ ma->ma_lmm_size = info->mti_big_lmmsize;
+ ma->ma_valid &= ~MA_LOV;
+ } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
+ ma->ma_lmv = info->mti_big_lmm;
+ ma->ma_lmv_size = info->mti_big_lmmsize;
+ ma->ma_valid &= ~MA_LMV;
+ } else {
+ LBUG();
+ }
+
+ LASSERT(!info->mti_big_lmm_used);
+ rc = __mdt_stripe_get(info, o, ma, name);
+ /* since big_lmm is always used here, clear 'used' flag to avoid
+ * assertion in mdt_big_xattr_get().
+ */
+ info->mti_big_lmm_used = 0;
+
+ return rc;
+}
+
int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
struct lu_fid *pfid)
{
RETURN(0);
}
+int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
+ struct lu_fid *pfid, struct lu_name *lname)
+{
+ struct lu_buf *buf = &info->mti_buf;
+ struct link_ea_header *leh;
+ struct link_ea_entry *lee;
+ int reclen;
+ int rc;
+
+ buf->lb_buf = info->mti_xattr_buf;
+ buf->lb_len = sizeof(info->mti_xattr_buf);
+ rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
+ XATTR_NAME_LINK);
+ if (rc == -ERANGE) {
+ rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
+ buf->lb_buf = info->mti_big_lmm;
+ buf->lb_len = info->mti_big_lmmsize;
+ }
+ if (rc < 0)
+ return rc;
+
+ if (rc < sizeof(*leh)) {
+ CERROR("short LinkEA on "DFID": rc = %d\n",
+ PFID(mdt_object_fid(o)), rc);
+ return -ENODATA;
+ }
+
+ leh = (struct link_ea_header *)buf->lb_buf;
+ lee = (struct link_ea_entry *)(leh + 1);
+ if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
+ leh->leh_magic = LINK_EA_MAGIC;
+ leh->leh_reccount = __swab32(leh->leh_reccount);
+ leh->leh_len = __swab64(leh->leh_len);
+ }
+ if (leh->leh_magic != LINK_EA_MAGIC)
+ return -EINVAL;
+
+ if (leh->leh_reccount == 0)
+ return -ENODATA;
+
+ linkea_entry_unpack(lee, &reclen, lname, pfid);
+
+ return 0;
+}
+
int mdt_attr_get_complex(struct mdt_thread_info *info,
struct mdt_object *o, struct md_attr *ma)
{
}
if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
- rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
+ rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
if (rc)
GOTO(out, rc);
}
if (need & MA_LMV && S_ISDIR(mode)) {
- rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
+ rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
if (rc != 0)
GOTO(out, rc);
}
if (need & MA_LMV_DEF && S_ISDIR(mode)) {
- rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
+ rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
if (rc != 0)
GOTO(out, rc);
}
}
static int mdt_getattr_internal(struct mdt_thread_info *info,
- struct mdt_object *o, int ma_need)
+ struct mdt_object *o, int ma_need)
{
- struct md_object *next = mdt_object_child(o);
- const struct mdt_body *reqbody = info->mti_body;
- struct ptlrpc_request *req = mdt_info_req(info);
- struct md_attr *ma = &info->mti_attr;
- struct lu_attr *la = &ma->ma_attr;
- struct req_capsule *pill = info->mti_pill;
- const struct lu_env *env = info->mti_env;
- struct mdt_body *repbody;
- struct lu_buf *buffer = &info->mti_buf;
- struct obd_export *exp = info->mti_exp;
- int rc;
+ struct mdt_device *mdt = info->mti_mdt;
+ struct md_object *next = mdt_object_child(o);
+ const struct mdt_body *reqbody = info->mti_body;
+ struct ptlrpc_request *req = mdt_info_req(info);
+ struct md_attr *ma = &info->mti_attr;
+ struct lu_attr *la = &ma->ma_attr;
+ struct req_capsule *pill = info->mti_pill;
+ const struct lu_env *env = info->mti_env;
+ struct mdt_body *repbody;
+ struct lu_buf *buffer = &info->mti_buf;
+ struct obd_export *exp = info->mti_exp;
+ ktime_t kstart = ktime_get();
+ int rc;
+
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
}
}
- if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
+ if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
reqbody->mbo_valid & OBD_MD_FLDIREA &&
- lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
- /* get default stripe info for this dir. */
- ma->ma_need |= MA_LOV_DEF;
- }
- ma->ma_need |= ma_need;
+ lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
+ /* get default stripe info for this dir. */
+ ma->ma_need |= MA_LOV_DEF;
+ }
+ ma->ma_need |= ma_need;
rc = mdt_attr_get_complex(info, o, ma);
if (unlikely(rc)) {
repbody->mbo_t_state = MS_RESTORE;
}
- if (likely(ma->ma_valid & MA_INODE))
- mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
- else
- RETURN(-EFAULT);
+ if (unlikely(!(ma->ma_valid & MA_INODE)))
+ RETURN(-EFAULT);
+
+ mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
- if (mdt_body_has_lov(la, reqbody)) {
- if (ma->ma_valid & MA_LOV) {
- LASSERT(ma->ma_lmm_size);
+ if (mdt_body_has_lov(la, reqbody)) {
+ u32 stripe_count = 1;
+
+ if (ma->ma_valid & MA_LOV) {
+ LASSERT(ma->ma_lmm_size);
repbody->mbo_eadatasize = ma->ma_lmm_size;
if (S_ISDIR(la->la_mode))
repbody->mbo_valid |= OBD_MD_FLDIREA;
else
repbody->mbo_valid |= OBD_MD_FLEASIZE;
mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
- }
+ }
if (ma->ma_valid & MA_LMV) {
+ struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
+ u32 magic = le32_to_cpu(lmv->lmv_magic);
+
/* Return -ENOTSUPP for old client */
if (!mdt_is_striped_client(req->rq_export))
RETURN(-ENOTSUPP);
mdt_dump_lmv(D_INFO, ma->ma_lmv);
repbody->mbo_eadatasize = ma->ma_lmv_size;
repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
+
+ stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
+ if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
+ mdt_restripe_migrate_add(info, o);
+ else if (magic == LMV_MAGIC_V1 &&
+ lmv_is_restriping(lmv))
+ mdt_restripe_update_add(info, o);
}
if (ma->ma_valid & MA_LMV_DEF) {
/* Return -ENOTSUPP for old client */
repbody->mbo_valid |= (OBD_MD_FLDIREA |
OBD_MD_DEFAULT_MEA);
}
+ CDEBUG(D_VFSTRACE,
+ "dirent count %llu stripe count %u MDT count %d\n",
+ ma->ma_attr.la_dirent_count, stripe_count,
+ atomic_read(&mdt->mdt_mds_mds_conns) + 1);
+ if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
+ ma->ma_attr.la_dirent_count >
+ mdt->mdt_restriper.mdr_dir_split_count &&
+ !fid_is_root(mdt_object_fid(o)) &&
+ mdt->mdt_enable_dir_auto_split &&
+ !o->mot_restriping &&
+ stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1)
+ mdt_auto_split_add(info, o);
} else if (S_ISLNK(la->la_mode) &&
reqbody->mbo_valid & OBD_MD_LINKNAME) {
buffer->lb_buf = ma->ma_lmm;
print_limit < rc ? "..." : "", print_limit,
(char *)ma->ma_lmm + rc - print_limit, rc);
rc = 0;
- }
- }
+ }
+ }
if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
#endif
out:
- if (rc == 0)
- mdt_counter_incr(req, LPROC_MDT_GETATTR);
+ if (rc == 0)
+ mdt_counter_incr(req, LPROC_MDT_GETATTR,
+ ktime_us_delta(ktime_get(), kstart));
- RETURN(rc);
+ RETURN(rc);
}
static int mdt_getattr(struct tgt_session_info *tsi)
__u64 child_bits,
struct ldlm_reply *ldlm_rep)
{
- struct ptlrpc_request *req = mdt_info_req(info);
- struct mdt_body *reqbody = NULL;
- struct mdt_object *parent = info->mti_object;
- struct mdt_object *child;
- struct lu_fid *child_fid = &info->mti_tmp_fid1;
- struct lu_name *lname = NULL;
+ struct ptlrpc_request *req = mdt_info_req(info);
+ struct mdt_body *reqbody = NULL;
+ struct mdt_object *parent = info->mti_object;
+ struct mdt_object *child = NULL;
+ struct lu_fid *child_fid = &info->mti_tmp_fid1;
+ struct lu_name *lname = NULL;
struct mdt_lock_handle *lhp = NULL;
- struct ldlm_lock *lock;
+ struct ldlm_lock *lock;
struct req_capsule *pill = info->mti_pill;
__u64 try_bits = 0;
bool is_resent;
mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
if (lu_name_is_valid(lname)) {
+ if (mdt_object_remote(parent)) {
+ CERROR("%s: parent "DFID" is on remote target\n",
+ mdt_obd_name(info->mti_mdt),
+ PFID(mdt_object_fid(parent)));
+ RETURN(-EPROTO);
+ }
+
CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
"ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
PNAME(lname), ldlm_rep);
RETURN(err_serious(-EPROTO));
*child_fid = reqbody->mbo_fid2;
-
if (unlikely(!fid_is_sane(child_fid)))
RETURN(err_serious(-EINVAL));
+ if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
+ mdt_object_get(info->mti_env, parent);
+ child = parent;
+ } else {
+ child = mdt_object_find(info->mti_env, info->mti_mdt,
+ child_fid);
+ if (IS_ERR(child))
+ RETURN(PTR_ERR(child));
+ }
+
+ if (mdt_object_remote(child)) {
+ CERROR("%s: child "DFID" is on remote target\n",
+ mdt_obd_name(info->mti_mdt),
+ PFID(mdt_object_fid(child)));
+ GOTO(out_child, rc = -EPROTO);
+ }
+
+ /* don't fetch LOOKUP lock if it's remote object */
+ rc = mdt_is_remote_object(info, parent, child);
+ if (rc < 0)
+ GOTO(out_child, rc);
+ if (rc)
+ child_bits &= ~MDS_INODELOCK_LOOKUP;
+
CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
"ldlm_rep = %p\n",
PFID(mdt_object_fid(parent)),
LU_OBJECT_DEBUG(D_INODE, info->mti_env,
&parent->mot_obj,
"Parent doesn't exist!");
- RETURN(-ESTALE);
- }
-
- if (mdt_object_remote(parent)) {
- CERROR("%s: parent "DFID" is on remote target\n",
- mdt_obd_name(info->mti_mdt),
- PFID(mdt_object_fid(parent)));
- RETURN(-EIO);
+ GOTO(out_child, rc = -ESTALE);
}
if (lu_name_is_valid(lname)) {
mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
if (rc != 0)
- GOTO(out_parent, rc);
- }
-
- mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
+ GOTO(unlock_parent, rc);
- /*
- *step 3: find the child object by fid & lock it.
- * regardless if it is local or remote.
- *
- *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
- * set parent dir fid the same as child fid in getattr by fid case
- * we should not lu_object_find() the object again, could lead
- * to hung if there is a concurrent unlink destroyed the object.
- */
- if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
- mdt_object_get(info->mti_env, parent);
- child = parent;
- } else {
child = mdt_object_find(info->mti_env, info->mti_mdt,
child_fid);
+ if (unlikely(IS_ERR(child)))
+ GOTO(unlock_parent, rc = PTR_ERR(child));
}
- if (unlikely(IS_ERR(child)))
- GOTO(out_parent, rc = PTR_ERR(child));
+ mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
+
+ /* step 3: lock child regardless if it is local or remote. */
+ LASSERT(child);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
if (!mdt_object_exists(child)) {
unlock_res_and_lock(lock);
}
LDLM_LOCK_PUT(lock);
- GOTO(out_parent, rc = 0);
+ GOTO(unlock_parent, rc = 0);
}
LDLM_LOCK_PUT(lock);
}
EXIT;
out_child:
- mdt_object_put(info->mti_env, child);
-out_parent:
+ if (child)
+ mdt_object_put(info->mti_env, child);
+unlock_parent:
if (lhp)
mdt_object_unlock(info, parent, lhp, 1);
return rc;
exp_max_brw_size(tsi->tsi_exp));
rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
PAGE_SHIFT;
- OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
- if (rdpg->rp_pages == NULL)
- RETURN(-ENOMEM);
+ OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
+ if (rdpg->rp_pages == NULL)
+ RETURN(-ENOMEM);
- for (i = 0; i < rdpg->rp_npages; ++i) {
+ for (i = 0; i < rdpg->rp_npages; ++i) {
rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
- if (rdpg->rp_pages[i] == NULL)
- GOTO(free_rdpg, rc = -ENOMEM);
- }
+ if (rdpg->rp_pages[i] == NULL)
+ GOTO(free_rdpg, rc = -ENOMEM);
+ }
- /* call lower layers to fill allocated pages with directory data */
+ /* call lower layers to fill allocated pages with directory data */
rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
if (rc < 0)
GOTO(free_rdpg, rc);
for (i = 0; i < rdpg->rp_npages; i++)
if (rdpg->rp_pages[i] != NULL)
__free_page(rdpg->rp_pages[i]);
- OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
+ OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
RETURN(0);
struct ptlrpc_request *req = tgt_ses_req(tsi);
struct req_capsule *pill = tsi->tsi_pill;
struct mdt_body *body;
+ ktime_t kstart = ktime_get();
int rc;
ENTRY;
mdt_thread_info_fini(info);
}
if (rc == 0)
- mdt_counter_incr(req, LPROC_MDT_SYNC);
+ mdt_counter_incr(req, LPROC_MDT_SYNC,
+ ktime_us_delta(ktime_get(), kstart));
RETURN(rc);
}
cache);
}
-static int mdt_object_local_lock(struct mdt_thread_info *info,
- struct mdt_object *o,
- struct mdt_lock_handle *lh, __u64 *ibits,
- __u64 trybits, bool cos_incompat)
+int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
+ struct mdt_lock_handle *lh, __u64 *ibits,
+ __u64 trybits, bool cos_incompat)
{
struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
union ldlm_policy_data *policy = &info->mti_policy;
next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
+
+ mdt_restriper_stop(m);
ping_evictor_stop();
/* Remove the HSM /proc entry so the coordinator cannot be
LASSERT(lsi->lsi_lmd);
/* CMD is supported only in IAM mode */
LASSERT(num);
- node_id = simple_strtol(num, NULL, 10);
+ rc = kstrtol(num, 10, &node_id);
+ if (rc)
+ RETURN(rc);
+
obd->u.obt.obt_magic = OBT_MAGIC;
if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
m->mdt_skip_lfsck = 1;
m->mdt_enable_remote_dir = 1;
m->mdt_enable_striped_dir = 1;
m->mdt_enable_dir_migration = 1;
+ m->mdt_enable_dir_restripe = 0;
+ m->mdt_enable_dir_auto_split = 0;
m->mdt_enable_remote_dir_gid = 0;
m->mdt_enable_chprojid_gid = 0;
m->mdt_enable_remote_rename = 1;
+ m->mdt_dir_restripe_nsonly = 1;
atomic_set(&m->mdt_mds_mds_conns, 0);
atomic_set(&m->mdt_async_commit_count, 0);
if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
m->mdt_lut.lut_local_recovery = 1;
+ rc = mdt_restriper_start(m);
+ if (rc)
+ GOTO(err_ping_evictor, rc);
+
RETURN(0);
+
+err_ping_evictor:
+ ping_evictor_stop();
err_procfs:
mdt_tunables_fini(m);
err_recovery:
init_rwsem(&mo->mot_dom_sem);
init_rwsem(&mo->mot_open_sem);
atomic_set(&mo->mot_open_count, 0);
+ mo->mot_restripe_offset = 0;
+ INIT_LIST_HEAD(&mo->mot_restripe_linkage);
RETURN(o);
}
RETURN(NULL);
RETURN(0);
}
+static inline void mdt_enable_slc(struct mdt_device *mdt)
+{
+ if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
+ mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
+}
+
+static inline void mdt_disable_slc(struct mdt_device *mdt)
+{
+ if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
+ mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
+}
+
/**
* Match client and server connection feature flags.
*
exp->exp_obd->obd_name, obd_export_nid2str(exp));
}
+ if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
+ !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
+ atomic_inc(&mdt->mdt_mds_mds_conns);
+ mdt_enable_slc(mdt);
+ }
+
return 0;
}
RETURN(rc);
}
-static inline void mdt_enable_slc(struct mdt_device *mdt)
-{
- if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
- mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
-}
-
-static inline void mdt_disable_slc(struct mdt_device *mdt)
-{
- if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
- mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
-}
-
static int mdt_obd_disconnect(struct obd_export *exp)
{
int rc;
mdt = mdt_dev(obd->obd_lu_dev);
- if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
- !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
- atomic_inc(&mdt->mdt_mds_mds_conns);
- mdt_enable_slc(mdt);
- }
-
/*
* first, check whether the stack is ready to handle requests
* XXX: probably not very appropriate method is used now
git://git.whamcloud.com / fs / lustre-release.git / blobdiff